Is artificial intelligence (AI) capable of flexible learning? Humans have the ability to learn from their situations and environments. In order for AI to do the same, it must first completely understand the learning process.
Unlike humans, AI has computational power that allows it to process copious amounts of data. As AI’s tasks continue to become more advanced, there’s also an increase in computational power necessary—which can become expensive. In order to avoid this need, AI was made to be a specific purpose learner. If AI gains the ability to relate data that is similar, it should be able to more efficiently process the data and understand it with a lower amount of power.
AI is already used in multiple disciplines and industries. These devices are capable of reading, writing, speaking, seeing, and understanding. They are able to recognize emotions, play games, and even debate. Due to these capabilities, AI is often leveraged as a virtual assistant that can help plan meetings; as a tool to simplify the job hunting and hiring processes; and much more.
However, in order for this technology to be able to advance, it needs to be able to learn.
Learning to Learn
To decrease the use of computational power, AI needs to determine and remember the most efficient learning path. After creating paths for various problems, AI can learn how to “multitask”, so that it may self-regulate and adjust its response based on the situation.
In order for AI to multitask, it needs to be equipped with the ability to analyze and correlate similar data sets. As it becomes more adaptive, AI will be able to apply learned knowledge to other situations.
It will take time for AI to become general-purpose learners like people. In order to achieve this, meta-reasoning and meta-learning are necessary. Meta-reasoning is about the efficient use of cognitive resources while meta-learning focuses on people’s unique ability to use limited cognitive resources and data to learn.
A key component to meta-reasoning is strategic thinking. This method relies on being able to see the big picture and make strategic choices. AI would have to be able to discover the strategic choices based on the situation and select from available strategies.
Meta-learning involves closing the gap between needing large amounts of data and using a small amount of data to learn. Some methods to do this include setting parameters that work for more than one task, defining a metric space, or simply mimicking surrounding.
Together, these two methods make up one a piece on the way to artificial intelligence becoming a Generalized Learner. Combining meta-reasoning and meta-learning with motor and sensory information will allow the machine to be more closely replicate human learning processes.
Applying Ethics to AI
How can we ethically make machines that resemble human likeness? Artificial Intelligence and Ethics in Design: Responsible Innovation is a two-part online training program from IEEE that focuses on integrating AI and autonomous systems within product and systems design.
Connect with an IEEE Specialist to learn more about bringing this program to your organization in order to better apply the ethics of AI to the business of design.
Interested in the course for yourself? Visit the IEEE Learning Network to access the individual course program.
Wu, Jun. (16 November 2019). Can Artificial Intelligence Learn to Learn? Forbes.
Marr, Bernard. (11 November 2019). 13 Mind-Blowing Things Artificial Intelligence Can Already Do Today. Forbes.